Question about genomic SEM

[kimd...@gmail.com]

Hello 

I am working on GWAS and i read your Genomic SEM methods article (Nature human behavior 2019) and and watched workshop at youtube with vivid interest.  

 

I'd really would appreciate your insight on a couple questions about genomic SEM. 

 

FIrst question is regarding the degree of freedom (df) obtained for Chi-squared when evaluating the model fit. When we use 3 summary statistics, we get df=0 which lead to NA values for chi-square and AIC. Would it be fine to evaluate the model fit solely based on CFI and SRMR? (I have attached the result obtained.)

In (Nature human behavior 2019), it states that 

"the number of freely estimated parameters does not exceed the number of non-redundant elements in the genetic covariance matrix being modelled" 

where I have interpreted it as there would be no problem if the number of freely estimated parameters equal to the number of non-redundant elements. 

Also, the example in Genomic SEM wiki explains a model generated regressing Educational Achievement on Bipolar Disorder and Schizophrenia is just a fully saturated model.

Another question is about dealing with GWAS summary statistics obtained for ordinal phenotype. In the Wiki, it has guidelines only for binary and continuous outcome.

Is it possible to conduct Genomic SEM for GWAS result of ordinal outcome?

According to the discussion (https://groups.google.com/g/genomic-sem-users/c/ffSDEt23GWc), you have mentioned that the the beta obtained should be converted. May I know how the conversion can take place? 

Another question is  about this error :  "Error in { : task 89417 failed - "Lapack routine dgesv: system is exactly singular: U[16,16] = 0" Calls: userGWAS -> %dopar% -> <Anonymous> Execution halted" at userGWAS() function. 

According to the discussion https://groups.google.com/g/genomic-sem-users/c/NZ0SwKS1luo/m/SQ6AXPFSBAAJ. you have suggested to use non-zero beta and SE sumstats but i can't found zero bata and SE variants. Could you check how to deal with this error?

 Last, i couldn't find a tools or software to represent results of Genomic SEM graphically as path diagrams. I've searched several tools like semplot in r package which draw a path diagrams of SEM, but still the figure on your articles looks more fancy. Could you check how did you draw path diagrams? 

 

Thank you for your time 

 

Sincerely and respectfully yours, 

 

Dan Say Kim 

[agro...@gmail.com]

Hi Dan, 

Answers below: 

For your first question, it's completely fine to run a fully saturated model with no df but you should not interpret model fit at all in this case (it's printing a CFI of 1 because the model perfectly fits the data). 

For ordinal outcomes there isn't a perfect answer here, but if you have more than 4 levels then I would suggest treating it as continuous and setting OLS = True when running the summary statistics function. 

For the error message about singular U[16,16] = 0 I would need some more information to diagnose what may be going on; can you post the code you ran that returned the error message?

For path diagrams unfortunately there isn't a program at the moment that will take GenomicSEM output and automatically generate the path diagrams. 

Best, 

  Andrew